Intelligent recommendation method and device for traditional Chinese medicine combination and electronic equipment

Abstract

The invention provides a traditional Chinese medicine combination intelligent recommendation method and device and electronic equipment, and the method comprises the steps: obtaining a to-be-consulted symptom description, and carrying out the semantic feature extraction of the to-be-consulted symptom description, and obtaining a symptom description feature corresponding to the to-be-consulted symptom description; a pre-trained traditional Chinese medicine combination recommendation model is called, the symptom description features are input into the traditional Chinese medicine combination recommendation model, a target traditional Chinese medicine combination corresponding to the symptom description features and output by the traditional Chinese medicine combination recommendation model is obtained, and the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function; the traditional Chinese medicine combination recommendation model is used for sequentially processing the symptom description features based on the residual network layer, the full connection layer and the activation function to obtain a target traditional Chinese medicine combination corresponding to the symptom description features. Automatic and accurate traditional Chinese medicine combination recommendation based on diseases is realized, the labor cost is reduced, and standardized and intelligent development of traditional Chinese medicine recommendation is promoted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traditional Chinese medicine combination recommendation, and in particular to a traditional Chinese medicine combination intelligent recommendation method and device and electronic equipment. BACKGROUND

[0002] According to related technologies, traditional Chinese medicine recommendation has long relied on expert experience and literature inheritance, and its diagnosis and prescription decision highly depends on subjective understanding of TCM theory and clinical experience accumulation of doctors. In addition, the traditional method does not fully utilize real-world TCM clinical data such as patient electronic medical records, and lacks fusion with modern clinical electronic medical record data, so that its prescription analysis lacks real adaptability and it is difficult to realize precise and personalized recommendation.

[0003] Therefore, finding an intelligent recommendation method capable of automatically and accurately recommending traditional Chinese medicine combination based on symptoms has become a current research hotspot. SUMMARY

[0004] The present application provides a traditional Chinese medicine combination intelligent recommendation method and device and electronic equipment, which realizes automatic and accurate traditional Chinese medicine combination recommendation based on symptoms, reduces labor cost, and promotes the standardization and intelligent development of traditional Chinese medicine recommendation.

[0005] The present application provides a traditional Chinese medicine combination intelligent recommendation method, which comprises: acquiring a to-be-consulted symptom description, and performing semantic feature extraction on the to-be-consulted symptom description to obtain a symptom description feature corresponding to the to-be-consulted symptom description; calling a pre-trained traditional Chinese medicine combination recommendation model, and inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is used to sequentially process the symptom description feature based on the residual network layer, the full connection layer and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature.

[0006] According to the method, the symptom description features are input into the traditional Chinese medicine combination recommendation model to obtain target traditional Chinese medicine combinations corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model.

[0007] According to the method, the symptom description features are input into the traditional Chinese medicine combination recommendation model to obtain target traditional Chinese medicine combinations corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model.

[0008] According to the method, the symptom description features are input into the traditional Chinese medicine combination recommendation model to obtain target traditional Chinese medicine combinations corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model.

[0009] According to the method, the symptom description features are input into the traditional Chinese medicine combination recommendation model to obtain target traditional Chinese medicine combinations corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model.

[0010] The application provides a traditional Chinese medicine combination intelligent recommendation method, which comprises the following steps: respectively performing semantic processing on historical symptom descriptions and historical traditional Chinese medicine combinations to obtain historical symptom description features and historical traditional Chinese medicine combination features, specifically comprising the following steps: writing a target script, wherein the target script is a script for batch processing of the semantic processing of historical symptom descriptions and historical traditional Chinese medicine combinations in multiple groups of historical clinical medical records; setting a target instruction, wherein the target instruction is a semantic processing requirement instruction for the semantic processing of historical symptom descriptions and historical traditional Chinese medicine combinations in multiple groups of historical clinical medical records during the running of the target script; and running the target script according to the target instruction to perform the semantic processing of historical symptom descriptions and historical traditional Chinese medicine combinations in multiple groups of historical clinical medical records, so as to obtain historical symptom description features and historical traditional Chinese medicine combination features.

[0011] According to the traditional Chinese medicine combination intelligent recommendation method provided by the application, after the training data is formed based on the historical symptom description features and the historical traditional Chinese medicine combination features, the method further comprises the following steps: performing mining processing on multiple traditional Chinese medicine combination labels in multiple training data based on an association rule mining algorithm to obtain frequent item sets in the multiple traditional Chinese medicine combination labels; forming optimized training data based on the frequent item sets and historical traditional Chinese medicine combination features corresponding to the frequent item sets, and forming an optimized training data set based on the optimized training data; and pre-training the traditional Chinese medicine combination recommendation model based on the training data set to obtain a trained traditional Chinese medicine combination recommendation model, specifically comprising the following steps: pre-training the traditional Chinese medicine combination recommendation model based on the optimized training data set to obtain a trained traditional Chinese medicine combination recommendation model.

[0012] According to the traditional Chinese medicine combination intelligent recommendation method provided by the application, the mining processing on the multiple traditional Chinese medicine combination labels in the multiple training data based on the association rule mining algorithm to obtain the frequent item sets in the multiple traditional Chinese medicine combination labels specifically comprises the following steps: setting a support threshold and / or a confidence threshold; and performing mining processing on the multiple traditional Chinese medicine combination labels in the multiple training data based on the association rule mining algorithm in combination with the support threshold and / or the confidence threshold to obtain the frequent item sets in the multiple traditional Chinese medicine combination labels.

[0013] The application further provides a traditional Chinese medicine combination intelligent recommendation device, the device comprising: an acquisition module, configured to acquire a symptom description to be consulted, and perform semantic feature extraction on the symptom description to be consulted to obtain a symptom description feature corresponding to the symptom description to be consulted; and a processing module, configured to call a pre-trained traditional Chinese medicine combination recommendation model, and input the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is configured to sequentially process the symptom description feature based on the residual network layer, the full connection layer and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature.

[0014] The application further provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the traditional Chinese medicine combination intelligent recommendation method according to any one of the above when executing the computer program.

[0015] The application further provides a non-transitory computer readable storage medium having a computer program stored thereon, and the computer program implements the traditional Chinese medicine combination intelligent recommendation method according to any one of the above when executed by a processor.

[0016] The application further provides a computer program product comprising a computer program, and the computer program implements the traditional Chinese medicine combination intelligent recommendation method according to any one of the above when executed by a processor.

[0017] The traditional Chinese medicine combination intelligent recommendation method, device and electronic device provided by the application, the method comprising: acquiring a symptom description to be consulted, and performing semantic feature extraction on the symptom description to be consulted to obtain a symptom description feature corresponding to the symptom description to be consulted; calling a pre-trained traditional Chinese medicine combination recommendation model, and inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is configured to sequentially process the symptom description feature based on the residual network layer, the full connection layer and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature. The method realizes automatic and accurate traditional Chinese medicine combination recommendation based on a disease, reduces the labor cost, and promotes the standardization and intelligent development of traditional Chinese medicine recommendation. BRIEF DESCRIPTION OF DRAWINGS

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is one of the flowcharts of the intelligent recommendation method for traditional Chinese medicine combinations provided by the present invention.

[0020] Figure 2 This is a flowchart illustrating the process of inputting the symptom description features into the traditional Chinese medicine combination recommendation model to obtain the target traditional Chinese medicine combination output by the traditional Chinese medicine combination recommendation model corresponding to the symptom description features.

[0021] Figure 3 This is a flowchart illustrating the process of generating training data based on historical clinical medical records, provided by the present invention.

[0022] Figure 4 This is a flowchart illustrating the process of semantically processing historical symptom descriptions and historical Chinese medicine combinations to obtain historical symptom description features and historical Chinese medicine combination features, as provided by the present invention.

[0023] Figure 5 This is a flowchart illustrating the process of pre-training the traditional Chinese medicine combination recommendation model based on the training dataset to obtain the trained traditional Chinese medicine combination recommendation model, as provided by the present invention.

[0024] Figure 6 This is a schematic diagram of the structure of the intelligent recommendation device for traditional Chinese medicine combinations provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] Figure 1 This is one of the flowcharts of the intelligent recommendation method for traditional Chinese medicine combinations provided by the present invention.

[0028] The following will combine Figure 1 The process of the intelligent recommendation method for traditional Chinese medicine combinations provided by this invention will be described.

[0029] In an example embodiment of the present application, in combination with Figure 1 It can be known that the intelligent recommendation method of traditional Chinese medicine combination can include steps 110 and 120, and each step will be introduced below.

[0030] In step 110, the symptom description to be consulted is obtained, and the semantic feature extraction of the symptom description to be consulted is performed to obtain the symptom description features corresponding to the symptom description to be consulted.

[0031] In an embodiment, the user-initiated symptom description to be consulted can be obtained, such as "repeated cough, yellow sputum, sticky sputum, red tongue, and thin yellow fur". Further, the semantic feature extraction of the symptom description to be consulted can be performed to obtain the symptom description features corresponding to the symptom description to be consulted. In an embodiment, the semantic feature extraction of the symptom description to be consulted can be performed based on a large model to generate corresponding symptom description features, for example, a 512-dimensional feature vector. It can be understood that the corresponding symptom description features can be the symptom description features of the same type of symptom description to be consulted. Through this embodiment, different symptom description methods can be simplified into a corresponding symptom description feature, which is convenient for obtaining the corresponding target traditional Chinese medicine combination based on the traditional Chinese medicine combination recommendation model.

[0032] In step 120, a pre-trained traditional Chinese medicine combination recommendation model is called, and the symptom description features are input into the traditional Chinese medicine combination recommendation model to obtain the target traditional Chinese medicine combination corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model.

[0033] The traditional Chinese medicine combination recommendation model includes a residual network layer, a full connection layer, and an activation function, and is used to sequentially process the symptom description features based on the residual network layer, the full connection layer, and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description features.

[0034] In an embodiment, the symptom description features can be input into the pre-trained traditional Chinese medicine combination recommendation model, so that the target traditional Chinese medicine combination corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model can be obtained. In the application process, the model sequentially performs the following processing: The residual network layer is used to perform nonlinear transformation and feature deepening on the input features to obtain feature encoding corresponding to the symptom description features, and to alleviate the gradient disappearance problem.

[0035] The full connection layer is used to map the deepened features (corresponding feature encoding) to a high-dimensional space to learn the complex association between symptoms and traditional Chinese medicine.

[0036] An activation function, such as a Sigmoid, is used to output the probability distribution of each traditional Chinese medicine combination, and then to determine the target traditional Chinese medicine combination based on the probability distribution of each traditional Chinese medicine combination.

[0037] It should be noted that the traditional Chinese medicine combination recommendation model provided by the application adopts a ResBlock module composed of double linear layers, integrates LeakyReLU and Dropout, effectively enhances the feature extraction and model generalization capability. Through the Sigmoid activation function, the multi-label prediction task of non-mutually exclusive labels can be supported, and the combination prediction problem of traditional Chinese medicine ingredients can be accurately modeled. This model structure realizes end-to-end training, has gradient stability and structural interpretability, and is significantly better than the application effect of traditional deep feedforward or convolutional models on structured medical data. Compared with the existing convolutional residual network model, the present model innovatively applies the residual structure to structured vector data while retaining its advantageous properties.

[0038] The specific structure of the Resblock module includes: two linear layers (Linear); two LeakyReLU activation functions; two Dropout layers (p=0.2) for preventing overfitting; and a residual connection path (IdentityMapping), also known as a skip connection, to realize gradient straight-through and alleviate the gradient vanishing problem in deep networks.

[0039] Among them, the ResBlock based on linear layer is more suitable for structured feature input in non-image (such as text, clinical, medicine, etc.) fields compared with the convolution structure used in the conventional ResNet. It is an innovation of structure transfer; Double LeakyReLU activation is more conducive to alleviating the "neuron death" problem and improving the non-linear modeling capability, especially in deep networks, compared with ReLU or single LeakyReLU; Dropout embedding, each layer integrates Dropout, effectively preventing overfitting, suitable for complex problems such as label space dense labels of traditional Chinese medicine; Constant dimension residual connection, since the input and output dimensions of the Linear layer are consistent, the skip connection does not need additional conversion, which is simple and efficient and preserves the gradient flow.

[0040] In the foregoing embodiments, through semantic feature extraction and deep feature extraction of the residual network, complex semantic information of the symptom description can be accurately captured, errors of traditional keyword matching can be avoided, and end-to-end mapping from high-dimensional features to traditional Chinese medicine combinations can be completed by the full connection layer and the activation function in cooperation, manual rule intervention can be avoided, and recommendation efficiency can be improved, so that automatic and accurate traditional Chinese medicine combination recommendation based on symptoms can be realized, manual cost is reduced, and the standardization and intelligent development of traditional Chinese medicine recommendation are promoted.

[0041] The intelligent traditional Chinese medicine combination recommendation method provided by the application comprises the following steps: obtaining a symptom description to be consulted, performing semantic feature extraction on the symptom description to be consulted, and obtaining a symptom description feature corresponding to the symptom description to be consulted; calling a pre-trained traditional Chinese medicine combination recommendation model, and inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is used for sequentially processing the symptom description feature based on the residual network layer, the full connection layer and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature. The method realizes automatic and accurate traditional Chinese medicine combination recommendation based on a disease, reduces labor cost, and promotes the standardization and intelligent development of traditional Chinese medicine recommendation.

[0042] Figure 2 FIG. 1 is a flowchart of a process of inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model provided by the application.

[0043] The application will be described in detail below with reference to the drawings. Figure 2 The process of inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model provided by the application will be described below.

[0044] In an exemplary embodiment of the application, the symptom description feature is input into the residual network layer in the traditional Chinese medicine combination recommendation model to obtain feature encoding corresponding to the symptom description feature output by the residual network layer. Figure 2 It can be known that the step of inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model can comprise steps 210 to 240, which will be introduced respectively.

[0045] In step 210, the symptom description feature is input into the residual network layer in the traditional Chinese medicine combination recommendation model to obtain feature encoding corresponding to the symptom description feature output by the residual network layer.

[0046] In an embodiment, the symptom description feature obtained by semantic feature extraction, such as the feature vectors of “headache-external wind cold syndrome” and “cough-wind dry lung syndrome”, can be input into the residual network layer. After residual network processing, a multi-dimensional feature encoding, for example, a 256-dimensional feature encoding, is output. The encoding can contain deep association information between symptoms and traditional Chinese medicine compatibility, such as the association strength of “external wind cold syndrome”, “obvious headache” and “obvious aversion to cold” with “ephedra” and “cassia twig”.

[0047] In an embodiment, the residual network layer can be composed of multiple (e.g., 2) residual blocks (ResBlock), wherein each residual block contains two linear layers (Linear) with input / output dimensions of 364 each, two LeakyReLU activation functions, and two Dropout layers (p = 0.2) to prevent overfitting.

[0048] In step 220, the feature encoding is input to a fully connected layer, and the feature encoding is mapped to a traditional Chinese medicine combination label space of a preset dimension based on the fully connected layer, and a traditional Chinese medicine combination label of the preset dimension corresponding to the feature encoding is output.

[0049] In an embodiment, the feature encoding output by the residual network can be input to a fully connected layer, and the number of neurons of the layer is preset to be the dimension of the traditional Chinese medicine combination label space (e.g., 1000 dimensions, corresponding to 1000 predefined traditional Chinese medicine combination labels). The fully connected layer maps the feature encoding to the label space through a weight matrix, and outputs a traditional Chinese medicine combination label of a preset dimension corresponding to the feature encoding.

[0050] In step 230, based on an activation function, the traditional Chinese medicine combination label of the preset dimension corresponding to the feature encoding is processed to obtain each probability value of the traditional Chinese medicine combination label of each preset dimension.

[0051] In step 240, based on each probability value of the traditional Chinese medicine combination label of each preset dimension, a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model is obtained.

[0052] In an embodiment, the Sigmoid activation function can be used to process the traditional Chinese medicine combination label of the preset dimension corresponding to the feature encoding output by the fully connected layer to obtain each probability value of the traditional Chinese medicine combination label of each preset dimension. For example, the probability of the "Mahuang" label is 0.72, the probability of the "Guizhi" label is 0.48, and the probability of the "Xinyi" label is 0.1. A binary decision module is constructed, and a threshold value θ = 0.5 is set as a classification boundary, and when the probability < 0.5, the value is assigned as 0, and when the probability ≥ 0.5, the value is assigned as 1. Further, the traditional Chinese medicine combination label with a probability value exceeding the set threshold value can be taken as the output result. Continue to use the previously described embodiment as an example for illustration, the probability value of "Mahuang" exceeds the set threshold value, and the value is assigned as 1, so the model can recommend the traditional Chinese medicine as the output.

[0053] In the foregoing embodiment, the gradient vanishing problem in deep network training is solved by residual structure, enabling the model to extract high-order correlation features of symptoms and traditional Chinese medicine combinations (such as the deep correspondence between the clinical manifestations of liver fire and headache, such as "bitter taste", "red eyes", and "rapid pulse" and the liver fire clearing drugs such as "gentiana scabra", "scrophularia ningpoensis", and "gardenia") and improve the richness and depth of feature expression. Furthermore, the label is converted into a probability value by the Sigmoid function, making the result interpretable (such as a probability value of 0.72 indicating that "ephedra" is suitable for prescribing in this symptom set).

[0054] In an exemplary embodiment of the present application, the foregoing embodiment is taken as an example for illustration. Based on the probability values of the traditional Chinese medicine combination labels under each preset dimension, the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model (corresponding to step 240) can be achieved in the following manner: Based on the probability values of the traditional Chinese medicine combination labels under each preset dimension, the traditional Chinese medicine combination labels with probability values exceeding a preset threshold under the preset dimension are obtained. Based on the traditional Chinese medicine combination labels with probability values exceeding a preset threshold under the preset dimension, the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model is obtained.

[0055] In an embodiment, the probability values of the traditional Chinese medicine combination labels under each preset dimension can be obtained after processing by an activation function (such as Sigmoid). For example, the model outputs a probability of 0.72 for "ephedra", a probability of 0.58 for "cassia twig", and a probability of 0.1 for "flos lonicerae". Since the probabilities of "ephedra" and "cassia twig" are greater than 0.5, the system outputs 1, indicating that they are recommended for use; the probability of "flos lonicerae" is <0.5, and the system outputs 0, indicating that it is not recommended for use. In this example, 0.5 can be used as the preset threshold, and it should be noted that the preset threshold can be adjusted according to actual conditions, which is not specifically limited in this embodiment. Further, the model outputs "ephedra" and "cassia twig" as the traditional Chinese medicine combination corresponding to the symptom set.

[0056] In another exemplary embodiment of the present application, the traditional Chinese medicine combination recommendation model can be pre-trained in the following manner: A training data set is constructed, wherein the training data set includes a plurality of training data, and the training data is formed based on historical clinical medical records; Based on the training data set, the traditional Chinese medicine combination recommendation model is pre-trained to obtain a trained traditional Chinese medicine combination recommendation model.

[0057] In an embodiment, the desensitized clinical medical record data can be extracted from a hospital information system (HIS), and medical records containing complete symptom descriptions and corresponding traditional Chinese medicine prescriptions are screened (such as "patient's symptoms: obvious headache, obvious chill, pale tongue, white fur, and floating and tight pulse; and modified medication: schizonepeta 9g, ledebouriella 6g, ephedra 6g, and cassia twig 3g"). The data is cleaned to remove noise (such as duplicate medical records and information missing records), and finally a training data set containing tens of thousands of structured data is formed, each data containing a symptom description text and a corresponding traditional Chinese medicine combination label, that is, the historical clinical medical record corresponding to the historical symptom description and the historical traditional Chinese medicine combination.

[0058] Further, the training data set can be input into the traditional Chinese medicine combination recommendation model and trained in a supervised learning manner. In the training process, the symptom description text can be input into the model, processed by the residual network layer and the full connection layer, and then the probability distribution of the traditional Chinese medicine combination label is output; then the cross-entropy loss function is used to calculate the difference between the model output probability and the real traditional Chinese medicine combination label; further, the model parameters (such as residual network layer weight and full connection layer bias) are adjusted by the gradient descent algorithm to minimize the loss value. In the application process, the above process can be repeated until the accuracy of the model on the validation set reaches the preset threshold (such as 95%), and finally the trained traditional Chinese medicine combination recommendation model is obtained. In the training process, the Batch Size can be set to 128; the Epoch can be set to 10; and the initial learning rate can be set to 0.01.

[0059] In yet another embodiment, five-fold cross-validation model can be used to verify the robustness and performance consistency of the model on different training data subsets. In each fold, 20% of the samples are divided as the validation set, and the rest are used for training. After iteration, the performance of each fold is summarized and analyzed (accuracy, precision, recall, F1 value, etc.).

[0060] In the foregoing embodiments, the training data set is constructed based on the real historical clinical medical record, which ensures that the model learns the clinically effective traditional Chinese medicine combination, such as the traditional Chinese medicine combination of "schizonepeta, ledebouriella, ephedra, and cassia twig" for the symptom set "obvious headache, obvious chill, pale tongue, white fur, and floating and tight pulse", which improves the clinical credibility of the recommendation result.

[0061] Figure 3 is a process diagram for forming training data based on historical clinical medical records provided by the application.

[0062] The following will be described in combination with Figure 3 The process of forming training data based on historical clinical medical records will be described.

[0063] In an exemplary embodiment of the application, the process of forming training data based on historical clinical medical records will be described in combination with Figure 3It can be seen that forming training data based on historical clinical medical records can include steps 310 to 330, which will be introduced respectively.

[0064] In step 310, a plurality of sets of historical clinical medical records are obtained, wherein the historical clinical medical records include historical symptom descriptions and historical traditional Chinese medicine combinations, and the historical traditional Chinese medicine combinations are used to treat the diseases corresponding to the historical symptom descriptions.

[0065] In an embodiment, the desensitized historical clinical medical record data is extracted from the medical institution database, and each medical record contains historical symptom descriptions (such as “ke down symptoms: cough, expectoration, yellow sputum, dry mouth”) and corresponding historical traditional Chinese medicine combinations (such as “addition and subtraction of drugs: mulberry leaf, chrysanthemum, apricot kernel, forsythia”).

[0066] In step 320, the historical symptom descriptions and the historical traditional Chinese medicine combinations are respectively subjected to semantic processing to obtain historical symptom description features and historical traditional Chinese medicine combination features.

[0067] In step 330, training data is formed based on the historical symptom description features and the historical traditional Chinese medicine combination features, wherein the historical traditional Chinese medicine combination features are used as the traditional Chinese medicine combination labels of the historical symptom description features.

[0068] Due to the diversity of expression and lack of standardized symptom descriptions in medical record texts, the historical symptom descriptions and the historical traditional Chinese medicine combinations can be respectively subjected to semantic processing to obtain historical symptom description features, that is, the historical symptom descriptions are subjected to semantic clustering and unified classification (such as “stomach discomfort”, “stomach fullness”, “stomach fullness” are unified and standardized as “stomach fullness discomfort”), and are automatically merged into 364 well-defined symptom feature items to form consistent input feature dimensions.

[0069] Based on the same principle, due to the names of traditional Chinese medicines appearing in the text, including aliases, abbreviations, writing errors, different processing forms, etc., the historical traditional Chinese medicine combinations can be subjected to semantic processing, context semantic discrimination, and automatic merging into unified named entities to obtain historical traditional Chinese medicine combination features. For example, “licorice (roasted)”, “roasted grass”, and “honey-roasted licorice” can be standardized as “roasted licorice”, achieving semantic unification and structure merging of 469 traditional Chinese medicine labels.

[0070] Further, the historical symptom description features and the corresponding historical traditional Chinese medicine combination features are paired to form structured training data. For example, symptom feature A is paired with traditional Chinese medicine combination feature X to form a piece of training data, wherein X is the traditional Chinese medicine combination label of A.

[0071] In an embodiment, the multi-threaded LLM API interface can also be invoked to perform entity extraction and semantic merging processes to automatically construct a structured input matrix (sample x 364 dimensions) and a label output matrix (sample x 469 dimensions) for subsequent model training.

[0072] In the present embodiment, the unstructured text medical records are converted into structured feature vectors through semantic processing, eliminating term differences and improving the consistency and computability of the training data. Furthermore, during the semantic processing, invalid information (such as the non-medical text "patient requests surgery as soon as possible" in the medical record) can be automatically identified and excluded, ensuring that the training data only contains features directly related to the symptom-TCM association and reducing the risk of model overfitting.

[0073] Figure 4 is a process diagram provided by the present application for respectively performing semantic processing on historical symptom descriptions and historical TCM combinations to obtain historical symptom description features and historical TCM combination features.

[0074] The following will be described in conjunction with Figure 4 The process of respectively performing semantic processing on historical symptom descriptions and historical TCM combinations to obtain historical symptom description features and historical TCM combination features provided by the present application will be described.

[0075] In an exemplary embodiment of the present application, in conjunction with Figure 4 It can be seen that respectively performing semantic processing on historical symptom descriptions and historical TCM combinations to obtain historical symptom description features and historical TCM combination features can include steps 410 to 430, which will be described below.

[0076] In step 410, a target script is written, wherein the target script is a script for batch processing of semantic processing of historical symptom descriptions and historical TCM combinations in multiple groups of historical clinical medical records.

[0077] In an embodiment, a batch processing script can be written in Python language, so that the following steps can be automatically performed: reading the desensitized historical clinical medical record data (in CSV or json format) through the Pandas library; calling a pre-trained Chinese medical semantic model (such as an improved model based on BERT) to perform word segmentation, entity recognition (such as extracting "cough" and "yellow sputum" as symptom entities) and relationship extraction on the symptom description. The script realizes batch processing of semantic processing of historical symptom descriptions and historical TCM combinations in multiple groups of historical clinical medical records. By integrating a TCM knowledge graph interface (such as the TCM-ID database), the preliminarily extracted TCM terms are subjected to semantic enhancement retrieval and comparison to improve the standardization, consistency and accuracy of professional expression after term merging.

[0078] In step 420, a target instruction is set, wherein the target instruction is a semantic processing requirement instruction for respectively performing semantic processing on the historical symptom descriptions and the historical traditional Chinese medicine combinations in the multiple sets of historical clinical medical records during running of the target script.

[0079] In step 430, the target script is called and run according to the target instruction, so as to respectively perform semantic processing on the historical symptom descriptions and the historical traditional Chinese medicine combinations in the multiple sets of historical clinical medical records, and obtain the historical symptom description features and the historical traditional Chinese medicine combination features.

[0080] In an embodiment, the semantic processing requirement instruction, that is, the target instruction, can be defined in the script, wherein the target instruction is a semantic processing requirement instruction for respectively performing semantic processing on the historical symptom descriptions and the historical traditional Chinese medicine combinations in the multiple sets of historical clinical medical records during running of the target script, for example, including performing named entity recognition on the symptom text, extracting symptoms and time entities, and labeling the correlation between the entities, such as “half a year ago-headache aggravation”.

[0081] Further, the script can be executed on a windows system according to the target instruction, and 100,000 pieces of historical medical record data can be processed in parallel through multiple processes. For example, 32 sets of medical records can be processed simultaneously in a single run, and after the symptom descriptions and the traditional Chinese medicine combinations of each set of medical records are respectively processed by a semantic model and a knowledge graph, a symptom feature vector with a dimension of 128 and a traditional Chinese medicine feature vector with a dimension of 64 are output, so as to obtain the historical symptom description features and the historical traditional Chinese medicine combination features. In this embodiment, the script is used to batch process medical record data, replacing manual annotation, and significantly reducing the labor cost of semantic processing.

[0082] Figure 5 is a flowchart of a process of pre-training the traditional Chinese medicine combination recommendation model based on the training data set and obtaining the trained traditional Chinese medicine combination recommendation model.

[0083] The following will be described in combination with Figure 5 The process of pre-training the traditional Chinese medicine combination recommendation model based on the training data set and obtaining the trained traditional Chinese medicine combination recommendation model provided by the present application will be described.

[0084] In an exemplary embodiment of the present application, the process of pre-training the traditional Chinese medicine combination recommendation model based on the training data set and obtaining the trained traditional Chinese medicine combination recommendation model is combined with Figure 5 It can be known that the process of pre-training the traditional Chinese medicine combination recommendation model based on the training data set and obtaining the trained traditional Chinese medicine combination recommendation model can include steps 510 to 530, and each step will be introduced below.

[0085] In step 510, based on an association rule mining algorithm, multiple traditional Chinese medicine combination labels in multiple training data are mined and processed to obtain frequent item sets in the multiple traditional Chinese medicine combination labels.

[0086] Real-world data exists in the high dimension of traditional Chinese medicine label, uneven frequency distribution, and mutual interference between labels. To alleviate these problems, the Apriori algorithm is introduced for label optimization and structure modeling.

[0087] In one embodiment, after forming the initial training data set, that is, the training data described above, a frequent item set mining algorithm (Apriori algorithm) can be used to mine the traditional Chinese medicine combination label. The minimum support threshold can be set to 0.1 (i.e., the combination with an appearance frequency of ≥10%). The frequent item set is extracted from 100,000 training data. For example, the support of the item set "Huangjing, Baishu" in the data set is 5%, which does not reach the threshold; while the support of the item set "Baishao, Fuling, Huangqin, Jinejin" in the data set is 12%, which is reserved as a frequent item set.

[0088] In step 520, based on the frequent item set and the historical traditional Chinese medicine combination characteristics corresponding to the frequent item set, an optimized training data is formed, and an optimized training data set is formed based on the optimized training data.

[0089] In step 530, based on the optimized training data set, the traditional Chinese medicine combination recommendation model is pre-trained to obtain a trained traditional Chinese medicine combination recommendation model.

[0090] In one embodiment, based on the mined frequent item set (such as "Danggui, Baishao" and "Laifensi, Houpu"), the records in the original training data that do not contain the frequent item set can be filtered, and the historical symptom description characteristics and the corresponding historical traditional Chinese medicine combination characteristics associated with the frequent item set are retained to form an optimized training data set (such as reduced from 100,000 to 80,000, but covering 95% of the high-frequency effective combination). Further, based on the optimized training data set, the traditional Chinese medicine combination recommendation model can be pre-trained to obtain a trained traditional Chinese medicine combination recommendation model. In this embodiment, the low-frequency traditional Chinese medicine combination label is filtered by association rule mining, the clinically common and effective combination is retained, the noise and redundancy in the training data are reduced, and the representativeness and practicability of the data set are improved.

[0091] In another exemplary embodiment of the present application, based on the association rule mining algorithm, a plurality of traditional Chinese medicine combination labels in a plurality of training data are mined to obtain a frequent item set in the plurality of traditional Chinese medicine combination labels, which can be implemented in the following way: Setting a support threshold and / or a confidence threshold; Based on the association rule mining algorithm, the support threshold and / or the confidence threshold are combined to mine a plurality of traditional Chinese medicine combination labels in a plurality of training data to obtain a frequent item set in the plurality of traditional Chinese medicine combination labels.

[0092] In one embodiment, a support threshold and / or a confidence threshold can be set; further, based on an association rule mining algorithm, combined with the support threshold and / or the confidence threshold, multiple traditional Chinese medicine combination labels in multiple training data can be mined to obtain frequent itemsets in multiple traditional Chinese medicine combination labels.

[0093] In application, the Apriori algorithm can be used to mine frequent itemsets (such as classic drug pairs and triplet formulas) in traditional Chinese medicine combinations; support and / or confidence thresholds can be set to select high-quality drug combination labels. In this embodiment, the Apriori mining results are matched with the neural network output labels to perform rule-level interpretation of the recommendation results, realizing a traditional Chinese medicine combination recommendation system with "traceable recommendation reasons." That is, the model is trained using frequent itemsets selected according to the support and / or confidence thresholds as training data, ensuring that the model's recommendation reasons are traceable.

[0094] In another embodiment, frequency analysis can be used to screen out low-frequency (usage frequency <10) Chinese medicine labels and retain a subset of labels with strong clinical representativeness to enhance the stability of model training.

[0095] As described above, the intelligent recommendation method for traditional Chinese medicine (TCM) combinations provided by this invention includes: acquiring a description of the symptoms to be consulted, and extracting semantic features from the description of the symptoms to obtain symptom description features corresponding to the description of the symptoms to be consulted; calling a pre-trained TCM combination recommendation model, and inputting the symptom description features into the TCM combination recommendation model to obtain the target TCM combination output by the TCM combination recommendation model corresponding to the symptom description features. The TCM combination recommendation model includes a residual network layer, a fully connected layer, and an activation function. The TCM combination recommendation model is used to process the symptom description features sequentially based on the residual network layer, the fully connected layer, and the activation function to obtain the target TCM combination corresponding to the symptom description features. This method achieves automatic and accurate TCM combination recommendation based on symptoms, reduces manual costs, and promotes the standardization and intelligent development of TCM recommendation.

[0096] The intelligent recommendation device for traditional Chinese medicine combinations provided by the present invention is described below. The intelligent recommendation device for traditional Chinese medicine combinations described below can be referred to in correspondence with the intelligent recommendation method for traditional Chinese medicine combinations described above.

[0097] Figure 6 This is a schematic diagram of the structure of the intelligent recommendation device for traditional Chinese medicine combinations provided by the present invention.

[0098] The following will combine Figure 6 The structure of the intelligent recommendation device for traditional Chinese medicine combinations provided by the present invention will be described.

[0099] In an example embodiment of the present application, in combination Figure 6 It can be known that the traditional Chinese medicine combination intelligent recommendation device can include an acquisition module 610 and a processing module 620, and each module will be introduced below.

[0100] The acquisition module 610 can be configured to acquire a symptom description to be consulted, and perform semantic feature extraction on the symptom description to be consulted to obtain a symptom description feature corresponding to the symptom description to be consulted; The processing module 620 can be configured to call a pre-trained traditional Chinese medicine combination recommendation model, and input the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein The traditional Chinese medicine combination recommendation model includes a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is used to sequentially process the symptom description feature based on the residual network layer, the full connection layer and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature.

[0101] In an example embodiment of the present application, the processing module 620 can implement the input of the symptom description feature into the traditional Chinese medicine combination recommendation model in the following manner to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model: The symptom description feature is input into the residual network layer in the traditional Chinese medicine combination recommendation model to obtain a feature code corresponding to the symptom description feature output by the residual network layer; The feature code is input into the full connection layer, and the feature code is mapped to a traditional Chinese medicine combination label space under a preset dimension based on the full connection layer to output a traditional Chinese medicine combination label under a preset dimension corresponding to the feature code; Based on the activation function, the traditional Chinese medicine combination label under the preset dimension corresponding to the feature code is processed to obtain each probability value of the traditional Chinese medicine combination label under each preset dimension; Based on each probability value of the traditional Chinese medicine combination label under each preset dimension, the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model is obtained.

[0102] In an example embodiment of the present application, the processing module 620 can implement the obtaining of the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model based on each probability value of the traditional Chinese medicine combination label under each preset dimension in the following manner: Based on each probability value of the traditional Chinese medicine combination label under each preset dimension, a traditional Chinese medicine combination label with a probability value exceeding a preset threshold under a preset dimension is obtained; obtain the target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model based on a probability value of a traditional Chinese medicine combination label exceeding a preset threshold in a preset dimension.

[0103] In an example embodiment of the present application, the processing module 620 can implement the pre-training of the traditional Chinese medicine combination recommendation model in the following manner: construct a training data set, wherein the training data set includes a plurality of training data, and the training data is formed based on historical clinical medical records; pre-train the traditional Chinese medicine combination recommendation model based on the training data set to obtain a trained traditional Chinese medicine combination recommendation model.

[0104] In an example embodiment of the present application, the processing module 620 can implement the formation of the training data based on historical clinical medical records in the following manner: obtain a plurality of sets of historical clinical medical records, wherein the historical clinical medical records include historical symptom descriptions and historical traditional Chinese medicine combinations, and the historical traditional Chinese medicine combinations are used to treat the conditions corresponding to the historical symptom descriptions; respectively perform semantic processing on the historical symptom descriptions and the historical traditional Chinese medicine combinations to obtain historical symptom description features and historical traditional Chinese medicine combination features; form the training data based on the historical symptom description features and the historical traditional Chinese medicine combination features, wherein the historical traditional Chinese medicine combination features are used as traditional Chinese medicine combination labels of the historical symptom description features.

[0105] In an example embodiment of the present application, the processing module 620 can implement the respective semantic processing of the historical symptom descriptions and the historical traditional Chinese medicine combinations to obtain the historical symptom description features and the historical traditional Chinese medicine combination features in the following manner: write a target script, wherein the target script is a script for batch processing of the respective semantic processing of the historical symptom descriptions and the historical traditional Chinese medicine combinations in the plurality of sets of historical clinical medical records; set a target instruction, wherein the target instruction is a semantic processing requirement instruction for the respective semantic processing of the historical symptom descriptions and the historical traditional Chinese medicine combinations in the plurality of sets of historical clinical medical records during the running of the target script; run the target script according to the target instruction to perform the respective semantic processing of the historical symptom descriptions and the historical traditional Chinese medicine combinations in the plurality of sets of historical clinical medical records, thereby obtaining the historical symptom description features and the historical traditional Chinese medicine combination features.

[0106] In an example embodiment of the present application, the processing module 620 can also be configured to: perform mining processing on a plurality of traditional Chinese medicine combination labels in a plurality of training data based on an association rule mining algorithm to obtain frequent item sets in the plurality of traditional Chinese medicine combination labels. forming an optimized training data set based on the optimized training data; The processing module 620 can implement the following manner to pre-train the traditional Chinese medicine combination recommendation model based on the training data set to obtain the trained traditional Chinese medicine combination recommendation model: The processing module 620 can implement the following manner to pre-train the traditional Chinese medicine combination recommendation model based on the training data set to obtain the trained traditional Chinese medicine combination recommendation model:

[0107] In an exemplary embodiment of the present application, the processing module 620 can also implement the following manner to mine the multiple traditional Chinese medicine combination labels in the multiple training data based on the association rule mining algorithm to obtain the frequent item set in the multiple traditional Chinese medicine combination labels: setting a support threshold and / or a confidence threshold; mining the multiple traditional Chinese medicine combination labels in the multiple training data based on the association rule mining algorithm in combination with the support threshold and / or the confidence threshold to obtain the frequent item set in the multiple traditional Chinese medicine combination labels.

[0108] Figure 7 An example of an entity structure schematic diagram of an electronic device is shown in Figure 7 As shown, the electronic device can include a processor 710, a communications interface 720, a memory 730, and a communications bus 740, wherein the processor 710, the communications interface 720, and the memory 730 complete mutual communication through the communications bus 740. The processor 710 can invoke a logical instruction in the memory 730 to execute a traditional Chinese medicine combination intelligent recommendation method, which includes: obtaining a to-be-consulted symptom description and performing semantic feature extraction on the to-be-consulted symptom description to obtain a symptom description feature corresponding to the to-be-consulted symptom description; invoking a pre-trained traditional Chinese medicine combination recommendation model and inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model includes a residual network layer, a full connection layer, and an activation function, the traditional Chinese medicine combination recommendation model is used to sequentially process the symptom description feature based on the residual network layer, the full connection layer, and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature.

[0109] In addition, the logic instructions in the memory 730 described above can be implemented in the form of software functional units and sold or used as independent products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0110] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor, so that the computer can execute the traditional Chinese medicine combination intelligent recommendation method provided by the above method. The method comprises the following steps: obtaining a to-be-consulted symptom description, and performing semantic feature extraction on the to-be-consulted symptom description to obtain a symptom description feature corresponding to the to-be-consulted symptom description; calling a pre-trained traditional Chinese medicine combination recommendation model, and inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is used to sequentially process the symptom description feature based on the residual network layer, the full connection layer and the activation function, to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature.

[0111] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a traditional Chinese medicine combination intelligent recommendation method provided by each of the above methods, and the method comprises: acquiring a to-be-consulted symptom description, and performing semantic feature extraction on the to-be-consulted symptom description to obtain a symptom description feature corresponding to the to-be-consulted symptom description; calling a pre-trained traditional Chinese medicine combination recommendation model, and inputting the symptom description feature into the traditional Chinese medicine combination recommendation model to obtain a target traditional Chinese medicine combination corresponding to the symptom description feature output by the traditional Chinese medicine combination recommendation model, wherein the traditional Chinese medicine combination recommendation model comprises a residual network layer, a full connection layer and an activation function, and the traditional Chinese medicine combination recommendation model is used to sequentially process the symptom description feature based on the residual network layer, the full connection layer and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description feature.

[0112] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0113] From the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platforms, and of course, it can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, server, or network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0114] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for intelligent recommendation of traditional Chinese medicine combinations, characterized in that, The method includes: Obtain the description of the symptom to be consulted, and extract semantic features from the description of the symptom to be consulted to obtain the symptom description features corresponding to the description of the symptom to be consulted; A pre-trained traditional Chinese medicine (TCM) combination recommendation model is invoked, and the symptom description features are input into the TCM combination recommendation model to obtain the target TCM combination output by the TCM combination recommendation model corresponding to the symptom description features. The traditional Chinese medicine combination recommendation model includes a residual network layer, a fully connected layer, and an activation function. The traditional Chinese medicine combination recommendation model is used to process the symptom description features sequentially based on the residual network layer, the fully connected layer, and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description features.

2. The intelligent recommendation method for traditional Chinese medicine combinations according to claim 1, characterized in that, The step of inputting the symptom description features into the traditional Chinese medicine combination recommendation model to obtain the target traditional Chinese medicine combination output by the traditional Chinese medicine combination recommendation model corresponding to the symptom description features specifically includes: The symptom description features are input into the residual network layer of the traditional Chinese medicine combination recommendation model to obtain the feature encoding output by the residual network layer corresponding to the symptom description features; The feature encoding is input into the fully connected layer, and the feature encoding is mapped to the Chinese herbal medicine combination label space under a preset dimension based on the fully connected layer, and the Chinese herbal medicine combination label under the preset dimension corresponding to the feature encoding is output. Based on the activation function, the traditional Chinese medicine combination labels under the preset dimensions corresponding to the feature encoding are processed to obtain the probability values ​​of the traditional Chinese medicine combination labels under each preset dimension. Based on the probability values ​​of each TCM combination label under each preset dimension, the target TCM combination output by the TCM combination recommendation model corresponding to the symptom description features is obtained.

3. The intelligent recommendation method for traditional Chinese medicine combinations according to claim 2, characterized in that, The method of obtaining the target traditional Chinese medicine combination corresponding to the symptom description features output by the traditional Chinese medicine combination recommendation model based on the probability values ​​of the traditional Chinese medicine combination tags under each preset dimension specifically includes: Based on the probability values ​​of each TCM combination label under each preset dimension, TCM combination labels with probability values ​​exceeding preset thresholds under each preset dimension are obtained. Based on the labels of traditional Chinese medicine combinations whose probability values ​​exceed a preset threshold under a preset dimension, the target traditional Chinese medicine combination output by the traditional Chinese medicine combination recommendation model is obtained, which corresponds to the symptom description features.

4. The intelligent recommendation method for traditional Chinese medicine combinations according to any one of claims 1 to 3, characterized in that, The traditional Chinese medicine combination recommendation model was pre-trained using the following method: Construct a training dataset, wherein the training dataset includes multiple training data, which are formed based on historical clinical medical records; Based on the training dataset, the traditional Chinese medicine combination recommendation model is pre-trained to obtain the trained traditional Chinese medicine combination recommendation model.

5. The intelligent recommendation method for traditional Chinese medicine combinations according to claim 4, characterized in that, The training data was generated based on historical clinical records in the following manner: Multiple sets of historical clinical medical records are obtained, wherein the historical clinical medical records include historical symptom descriptions and historical Chinese medicine combinations, wherein the historical Chinese medicine combinations are used to treat the diseases corresponding to the historical symptom descriptions; Semantic processing was performed on historical symptom descriptions and historical Chinese medicine combinations to obtain features of historical symptom descriptions and features of historical Chinese medicine combinations. The training data is formed based on the historical symptom description features and the historical Chinese medicine combination features, wherein the historical Chinese medicine combination features serve as the Chinese medicine combination labels for the historical symptom description features.

6. The intelligent recommendation method for traditional Chinese medicine combinations according to claim 5, characterized in that, The semantic processing of historical symptom descriptions and historical Chinese medicine combinations, respectively, to obtain historical symptom description features and historical Chinese medicine combination features, specifically includes: Write a target script, wherein the target script is a script that performs semantic processing on historical symptom descriptions and historical Chinese medicine combinations in multiple sets of historical clinical medical records in batches; Set target instructions, wherein the target instructions are semantic processing requirements instructions for semantic processing of historical symptom descriptions and historical Chinese medicine combinations in multiple sets of historical clinical medical records during the execution of the target script; According to the target instructions, the target script is invoked and executed to perform semantic processing on historical symptom descriptions and historical Chinese medicine combinations in multiple sets of historical clinical medical records, so as to obtain historical symptom description features and historical Chinese medicine combination features.

7. The intelligent recommendation method for traditional Chinese medicine combinations according to claim 5, characterized in that, After forming the training data based on the historical symptom description features and the historical traditional Chinese medicine combination features, the method further includes: Based on the association rule mining algorithm, multiple Chinese medicine combination labels in multiple training data are mined to obtain frequent itemsets in multiple Chinese medicine combination labels. Based on the frequent itemsets and the historical Chinese medicine combination features corresponding to the frequent itemsets, optimized training data is formed, and an optimized training dataset is formed based on the optimized training data. The step of pre-training the traditional Chinese medicine combination recommendation model based on the training dataset to obtain the trained traditional Chinese medicine combination recommendation model specifically includes: Based on the optimized training dataset, the traditional Chinese medicine combination recommendation model is pre-trained to obtain the trained traditional Chinese medicine combination recommendation model.

8. The intelligent recommendation method for traditional Chinese medicine combinations according to claim 7, characterized in that, The association rule mining algorithm processes multiple traditional Chinese medicine combination tags in the training data to obtain frequent itemsets among the multiple traditional Chinese medicine combination tags, specifically including: Set support thresholds and / or confidence thresholds; Based on the association rule mining algorithm, combined with support threshold and / or confidence threshold, multiple traditional Chinese medicine combination labels in multiple training data are mined to obtain frequent itemsets in multiple traditional Chinese medicine combination labels.

9. A smart recommendation device for traditional Chinese medicine combinations, characterized in that, The device includes: The acquisition module is used to acquire the description of the symptoms to be consulted, and to extract semantic features from the description of the symptoms to be consulted to obtain the symptom description features corresponding to the description of the symptoms to be consulted. The processing module is used to call a pre-trained traditional Chinese medicine (TCM) combination recommendation model, input the symptom description features into the TCM combination recommendation model, and obtain the target TCM combination output by the TCM combination recommendation model corresponding to the symptom description features. The traditional Chinese medicine combination recommendation model includes a residual network layer, a fully connected layer, and an activation function. The traditional Chinese medicine combination recommendation model is used to process the symptom description features sequentially based on the residual network layer, the fully connected layer, and the activation function to obtain the target traditional Chinese medicine combination corresponding to the symptom description features.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the intelligent recommendation method for traditional Chinese medicine combinations as described in any one of claims 1 to 8.

Images